1 /* 2 * Copyright (c) 1997, 2024, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #include "precompiled.hpp" 26 #include "compiler/compiler_globals.hpp" 27 #include "interp_masm_x86.hpp" 28 #include "interpreter/interpreter.hpp" 29 #include "interpreter/interpreterRuntime.hpp" 30 #include "logging/log.hpp" 31 #include "oops/arrayOop.hpp" 32 #include "oops/markWord.hpp" 33 #include "oops/methodData.hpp" 34 #include "oops/method.hpp" 35 #include "oops/resolvedFieldEntry.hpp" 36 #include "oops/resolvedIndyEntry.hpp" 37 #include "oops/resolvedMethodEntry.hpp" 38 #include "prims/jvmtiExport.hpp" 39 #include "prims/jvmtiThreadState.hpp" 40 #include "runtime/basicLock.hpp" 41 #include "runtime/frame.inline.hpp" 42 #include "runtime/javaThread.hpp" 43 #include "runtime/safepointMechanism.hpp" 44 #include "runtime/sharedRuntime.hpp" 45 #include "utilities/powerOfTwo.hpp" 46 47 // Implementation of InterpreterMacroAssembler 48 49 void InterpreterMacroAssembler::jump_to_entry(address entry) { 50 assert(entry, "Entry must have been generated by now"); 51 jump(RuntimeAddress(entry)); 52 } 53 54 void InterpreterMacroAssembler::profile_obj_type(Register obj, const Address& mdo_addr) { 55 Label update, next, none; 56 57 #ifdef _LP64 58 assert_different_registers(obj, rscratch1, mdo_addr.base(), mdo_addr.index()); 59 #else 60 assert_different_registers(obj, mdo_addr.base(), mdo_addr.index()); 61 #endif 62 63 interp_verify_oop(obj, atos); 64 65 testptr(obj, obj); 66 jccb(Assembler::notZero, update); 67 testptr(mdo_addr, TypeEntries::null_seen); 68 jccb(Assembler::notZero, next); // null already seen. Nothing to do anymore. 69 // atomic update to prevent overwriting Klass* with 0 70 lock(); 71 orptr(mdo_addr, TypeEntries::null_seen); 72 jmpb(next); 73 74 bind(update); 75 load_klass(obj, obj, rscratch1); 76 #ifdef _LP64 77 mov(rscratch1, obj); 78 #endif 79 80 xorptr(obj, mdo_addr); 81 testptr(obj, TypeEntries::type_klass_mask); 82 jccb(Assembler::zero, next); // klass seen before, nothing to 83 // do. The unknown bit may have been 84 // set already but no need to check. 85 86 testptr(obj, TypeEntries::type_unknown); 87 jccb(Assembler::notZero, next); // already unknown. Nothing to do anymore. 88 89 cmpptr(mdo_addr, 0); 90 jccb(Assembler::equal, none); 91 cmpptr(mdo_addr, TypeEntries::null_seen); 92 jccb(Assembler::equal, none); 93 #ifdef _LP64 94 // There is a chance that the checks above (re-reading profiling 95 // data from memory) fail if another thread has just set the 96 // profiling to this obj's klass 97 mov(obj, rscratch1); 98 xorptr(obj, mdo_addr); 99 testptr(obj, TypeEntries::type_klass_mask); 100 jccb(Assembler::zero, next); 101 #endif 102 103 // different than before. Cannot keep accurate profile. 104 orptr(mdo_addr, TypeEntries::type_unknown); 105 jmpb(next); 106 107 bind(none); 108 // first time here. Set profile type. 109 movptr(mdo_addr, obj); 110 #ifdef ASSERT 111 andptr(obj, TypeEntries::type_klass_mask); 112 verify_klass_ptr(obj); 113 #endif 114 115 bind(next); 116 } 117 118 void InterpreterMacroAssembler::profile_arguments_type(Register mdp, Register callee, Register tmp, bool is_virtual) { 119 if (!ProfileInterpreter) { 120 return; 121 } 122 123 if (MethodData::profile_arguments() || MethodData::profile_return()) { 124 Label profile_continue; 125 126 test_method_data_pointer(mdp, profile_continue); 127 128 int off_to_start = is_virtual ? in_bytes(VirtualCallData::virtual_call_data_size()) : in_bytes(CounterData::counter_data_size()); 129 130 cmpb(Address(mdp, in_bytes(DataLayout::tag_offset()) - off_to_start), is_virtual ? DataLayout::virtual_call_type_data_tag : DataLayout::call_type_data_tag); 131 jcc(Assembler::notEqual, profile_continue); 132 133 if (MethodData::profile_arguments()) { 134 Label done; 135 int off_to_args = in_bytes(TypeEntriesAtCall::args_data_offset()); 136 addptr(mdp, off_to_args); 137 138 for (int i = 0; i < TypeProfileArgsLimit; i++) { 139 if (i > 0 || MethodData::profile_return()) { 140 // If return value type is profiled we may have no argument to profile 141 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args)); 142 subl(tmp, i*TypeStackSlotEntries::per_arg_count()); 143 cmpl(tmp, TypeStackSlotEntries::per_arg_count()); 144 jcc(Assembler::less, done); 145 } 146 movptr(tmp, Address(callee, Method::const_offset())); 147 load_unsigned_short(tmp, Address(tmp, ConstMethod::size_of_parameters_offset())); 148 // stack offset o (zero based) from the start of the argument 149 // list, for n arguments translates into offset n - o - 1 from 150 // the end of the argument list 151 subptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::stack_slot_offset(i))-off_to_args)); 152 subl(tmp, 1); 153 Address arg_addr = argument_address(tmp); 154 movptr(tmp, arg_addr); 155 156 Address mdo_arg_addr(mdp, in_bytes(TypeEntriesAtCall::argument_type_offset(i))-off_to_args); 157 profile_obj_type(tmp, mdo_arg_addr); 158 159 int to_add = in_bytes(TypeStackSlotEntries::per_arg_size()); 160 addptr(mdp, to_add); 161 off_to_args += to_add; 162 } 163 164 if (MethodData::profile_return()) { 165 movptr(tmp, Address(mdp, in_bytes(TypeEntriesAtCall::cell_count_offset())-off_to_args)); 166 subl(tmp, TypeProfileArgsLimit*TypeStackSlotEntries::per_arg_count()); 167 } 168 169 bind(done); 170 171 if (MethodData::profile_return()) { 172 // We're right after the type profile for the last 173 // argument. tmp is the number of cells left in the 174 // CallTypeData/VirtualCallTypeData to reach its end. Non null 175 // if there's a return to profile. 176 assert(ReturnTypeEntry::static_cell_count() < TypeStackSlotEntries::per_arg_count(), "can't move past ret type"); 177 shll(tmp, log2i_exact((int)DataLayout::cell_size)); 178 addptr(mdp, tmp); 179 } 180 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp); 181 } else { 182 assert(MethodData::profile_return(), "either profile call args or call ret"); 183 update_mdp_by_constant(mdp, in_bytes(TypeEntriesAtCall::return_only_size())); 184 } 185 186 // mdp points right after the end of the 187 // CallTypeData/VirtualCallTypeData, right after the cells for the 188 // return value type if there's one 189 190 bind(profile_continue); 191 } 192 } 193 194 void InterpreterMacroAssembler::profile_return_type(Register mdp, Register ret, Register tmp) { 195 assert_different_registers(mdp, ret, tmp, _bcp_register); 196 if (ProfileInterpreter && MethodData::profile_return()) { 197 Label profile_continue; 198 199 test_method_data_pointer(mdp, profile_continue); 200 201 if (MethodData::profile_return_jsr292_only()) { 202 assert(Method::intrinsic_id_size_in_bytes() == 2, "assuming Method::_intrinsic_id is u2"); 203 204 // If we don't profile all invoke bytecodes we must make sure 205 // it's a bytecode we indeed profile. We can't go back to the 206 // beginning of the ProfileData we intend to update to check its 207 // type because we're right after it and we don't known its 208 // length 209 Label do_profile; 210 cmpb(Address(_bcp_register, 0), Bytecodes::_invokedynamic); 211 jcc(Assembler::equal, do_profile); 212 cmpb(Address(_bcp_register, 0), Bytecodes::_invokehandle); 213 jcc(Assembler::equal, do_profile); 214 get_method(tmp); 215 cmpw(Address(tmp, Method::intrinsic_id_offset()), static_cast<int>(vmIntrinsics::_compiledLambdaForm)); 216 jcc(Assembler::notEqual, profile_continue); 217 218 bind(do_profile); 219 } 220 221 Address mdo_ret_addr(mdp, -in_bytes(ReturnTypeEntry::size())); 222 mov(tmp, ret); 223 profile_obj_type(tmp, mdo_ret_addr); 224 225 bind(profile_continue); 226 } 227 } 228 229 void InterpreterMacroAssembler::profile_parameters_type(Register mdp, Register tmp1, Register tmp2) { 230 if (ProfileInterpreter && MethodData::profile_parameters()) { 231 Label profile_continue; 232 233 test_method_data_pointer(mdp, profile_continue); 234 235 // Load the offset of the area within the MDO used for 236 // parameters. If it's negative we're not profiling any parameters 237 movl(tmp1, Address(mdp, in_bytes(MethodData::parameters_type_data_di_offset()) - in_bytes(MethodData::data_offset()))); 238 testl(tmp1, tmp1); 239 jcc(Assembler::negative, profile_continue); 240 241 // Compute a pointer to the area for parameters from the offset 242 // and move the pointer to the slot for the last 243 // parameters. Collect profiling from last parameter down. 244 // mdo start + parameters offset + array length - 1 245 addptr(mdp, tmp1); 246 movptr(tmp1, Address(mdp, ArrayData::array_len_offset())); 247 decrement(tmp1, TypeStackSlotEntries::per_arg_count()); 248 249 Label loop; 250 bind(loop); 251 252 int off_base = in_bytes(ParametersTypeData::stack_slot_offset(0)); 253 int type_base = in_bytes(ParametersTypeData::type_offset(0)); 254 Address::ScaleFactor per_arg_scale = Address::times(DataLayout::cell_size); 255 Address arg_off(mdp, tmp1, per_arg_scale, off_base); 256 Address arg_type(mdp, tmp1, per_arg_scale, type_base); 257 258 // load offset on the stack from the slot for this parameter 259 movptr(tmp2, arg_off); 260 negptr(tmp2); 261 // read the parameter from the local area 262 movptr(tmp2, Address(_locals_register, tmp2, Interpreter::stackElementScale())); 263 264 // profile the parameter 265 profile_obj_type(tmp2, arg_type); 266 267 // go to next parameter 268 decrement(tmp1, TypeStackSlotEntries::per_arg_count()); 269 jcc(Assembler::positive, loop); 270 271 bind(profile_continue); 272 } 273 } 274 275 void InterpreterMacroAssembler::call_VM_leaf_base(address entry_point, 276 int number_of_arguments) { 277 // interpreter specific 278 // 279 // Note: No need to save/restore bcp & locals registers 280 // since these are callee saved registers and no blocking/ 281 // GC can happen in leaf calls. 282 // Further Note: DO NOT save/restore bcp/locals. If a caller has 283 // already saved them so that it can use rsi/rdi as temporaries 284 // then a save/restore here will DESTROY the copy the caller 285 // saved! There used to be a save_bcp() that only happened in 286 // the ASSERT path (no restore_bcp). Which caused bizarre failures 287 // when jvm built with ASSERTs. 288 #ifdef ASSERT 289 { 290 Label L; 291 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD); 292 jcc(Assembler::equal, L); 293 stop("InterpreterMacroAssembler::call_VM_leaf_base:" 294 " last_sp != null"); 295 bind(L); 296 } 297 #endif 298 // super call 299 MacroAssembler::call_VM_leaf_base(entry_point, number_of_arguments); 300 // interpreter specific 301 // LP64: Used to ASSERT that r13/r14 were equal to frame's bcp/locals 302 // but since they may not have been saved (and we don't want to 303 // save them here (see note above) the assert is invalid. 304 } 305 306 void InterpreterMacroAssembler::call_VM_base(Register oop_result, 307 Register java_thread, 308 Register last_java_sp, 309 address entry_point, 310 int number_of_arguments, 311 bool check_exceptions) { 312 // interpreter specific 313 // 314 // Note: Could avoid restoring locals ptr (callee saved) - however doesn't 315 // really make a difference for these runtime calls, since they are 316 // slow anyway. Btw., bcp must be saved/restored since it may change 317 // due to GC. 318 NOT_LP64(assert(java_thread == noreg , "not expecting a precomputed java thread");) 319 save_bcp(); 320 #ifdef ASSERT 321 { 322 Label L; 323 cmpptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), NULL_WORD); 324 jcc(Assembler::equal, L); 325 stop("InterpreterMacroAssembler::call_VM_base:" 326 " last_sp isn't null"); 327 bind(L); 328 } 329 #endif /* ASSERT */ 330 // super call 331 MacroAssembler::call_VM_base(oop_result, noreg, last_java_sp, 332 entry_point, number_of_arguments, 333 check_exceptions); 334 // interpreter specific 335 restore_bcp(); 336 restore_locals(); 337 } 338 339 #ifdef _LP64 340 void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result, 341 address entry_point, 342 Register arg_1) { 343 assert(arg_1 == c_rarg1, ""); 344 Label resume_pc, not_preempted; 345 346 #ifdef ASSERT 347 { 348 Label L; 349 cmpptr(Address(r15_thread, JavaThread::preempt_alternate_return_offset()), NULL_WORD); 350 jcc(Assembler::equal, L); 351 stop("Should not have alternate return address set"); 352 bind(L); 353 } 354 #endif /* ASSERT */ 355 356 push_cont_fastpath(); 357 358 // Make VM call. In case of preemption set last_pc to the one we want to resume to. 359 lea(rscratch1, resume_pc); 360 push(rscratch1); 361 MacroAssembler::call_VM_helper(oop_result, entry_point, 1, false /*check_exceptions*/); 362 pop(rscratch1); 363 364 pop_cont_fastpath(); 365 366 // Check if preempted. 367 movptr(rscratch1, Address(r15_thread, JavaThread::preempt_alternate_return_offset())); 368 cmpptr(rscratch1, NULL_WORD); 369 jccb(Assembler::zero, not_preempted); 370 movptr(Address(r15_thread, JavaThread::preempt_alternate_return_offset()), NULL_WORD); 371 jmp(rscratch1); 372 373 // In case of preemption, this is where we will resume once we finally acquire the monitor. 374 bind(resume_pc); 375 restore_after_resume(false /* is_native */); 376 377 bind(not_preempted); 378 } 379 380 void InterpreterMacroAssembler::restore_after_resume(bool is_native) { 381 lea(rscratch1, ExternalAddress(Interpreter::cont_resume_interpreter_adapter())); 382 call(rscratch1); 383 if (is_native) { 384 // On resume we need to set up stack as expected. 385 push(dtos); 386 push(ltos); 387 } 388 } 389 #else 390 void InterpreterMacroAssembler::call_VM_preemptable(Register oop_result, 391 address entry_point, 392 Register arg_1) { 393 MacroAssembler::call_VM(oop_result, entry_point, arg_1); 394 } 395 #endif // _LP64 396 397 void InterpreterMacroAssembler::check_and_handle_popframe(Register java_thread) { 398 if (JvmtiExport::can_pop_frame()) { 399 Label L; 400 // Initiate popframe handling only if it is not already being 401 // processed. If the flag has the popframe_processing bit set, it 402 // means that this code is called *during* popframe handling - we 403 // don't want to reenter. 404 // This method is only called just after the call into the vm in 405 // call_VM_base, so the arg registers are available. 406 Register pop_cond = NOT_LP64(java_thread) // Not clear if any other register is available on 32 bit 407 LP64_ONLY(c_rarg0); 408 movl(pop_cond, Address(java_thread, JavaThread::popframe_condition_offset())); 409 testl(pop_cond, JavaThread::popframe_pending_bit); 410 jcc(Assembler::zero, L); 411 testl(pop_cond, JavaThread::popframe_processing_bit); 412 jcc(Assembler::notZero, L); 413 // Call Interpreter::remove_activation_preserving_args_entry() to get the 414 // address of the same-named entrypoint in the generated interpreter code. 415 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_preserving_args_entry)); 416 jmp(rax); 417 bind(L); 418 NOT_LP64(get_thread(java_thread);) 419 } 420 } 421 422 void InterpreterMacroAssembler::load_earlyret_value(TosState state) { 423 Register thread = LP64_ONLY(r15_thread) NOT_LP64(rcx); 424 NOT_LP64(get_thread(thread);) 425 movptr(rcx, Address(thread, JavaThread::jvmti_thread_state_offset())); 426 const Address tos_addr(rcx, JvmtiThreadState::earlyret_tos_offset()); 427 const Address oop_addr(rcx, JvmtiThreadState::earlyret_oop_offset()); 428 const Address val_addr(rcx, JvmtiThreadState::earlyret_value_offset()); 429 #ifdef _LP64 430 switch (state) { 431 case atos: movptr(rax, oop_addr); 432 movptr(oop_addr, NULL_WORD); 433 interp_verify_oop(rax, state); break; 434 case ltos: movptr(rax, val_addr); break; 435 case btos: // fall through 436 case ztos: // fall through 437 case ctos: // fall through 438 case stos: // fall through 439 case itos: movl(rax, val_addr); break; 440 case ftos: load_float(val_addr); break; 441 case dtos: load_double(val_addr); break; 442 case vtos: /* nothing to do */ break; 443 default : ShouldNotReachHere(); 444 } 445 // Clean up tos value in the thread object 446 movl(tos_addr, ilgl); 447 movl(val_addr, NULL_WORD); 448 #else 449 const Address val_addr1(rcx, JvmtiThreadState::earlyret_value_offset() 450 + in_ByteSize(wordSize)); 451 switch (state) { 452 case atos: movptr(rax, oop_addr); 453 movptr(oop_addr, NULL_WORD); 454 interp_verify_oop(rax, state); break; 455 case ltos: 456 movl(rdx, val_addr1); // fall through 457 case btos: // fall through 458 case ztos: // fall through 459 case ctos: // fall through 460 case stos: // fall through 461 case itos: movl(rax, val_addr); break; 462 case ftos: load_float(val_addr); break; 463 case dtos: load_double(val_addr); break; 464 case vtos: /* nothing to do */ break; 465 default : ShouldNotReachHere(); 466 } 467 #endif // _LP64 468 // Clean up tos value in the thread object 469 movl(tos_addr, ilgl); 470 movptr(val_addr, NULL_WORD); 471 NOT_LP64(movptr(val_addr1, NULL_WORD);) 472 } 473 474 475 void InterpreterMacroAssembler::check_and_handle_earlyret(Register java_thread) { 476 if (JvmtiExport::can_force_early_return()) { 477 Label L; 478 Register tmp = LP64_ONLY(c_rarg0) NOT_LP64(java_thread); 479 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(java_thread); 480 481 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset())); 482 testptr(tmp, tmp); 483 jcc(Assembler::zero, L); // if (thread->jvmti_thread_state() == nullptr) exit; 484 485 // Initiate earlyret handling only if it is not already being processed. 486 // If the flag has the earlyret_processing bit set, it means that this code 487 // is called *during* earlyret handling - we don't want to reenter. 488 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_state_offset())); 489 cmpl(tmp, JvmtiThreadState::earlyret_pending); 490 jcc(Assembler::notEqual, L); 491 492 // Call Interpreter::remove_activation_early_entry() to get the address of the 493 // same-named entrypoint in the generated interpreter code. 494 NOT_LP64(get_thread(java_thread);) 495 movptr(tmp, Address(rthread, JavaThread::jvmti_thread_state_offset())); 496 #ifdef _LP64 497 movl(tmp, Address(tmp, JvmtiThreadState::earlyret_tos_offset())); 498 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), tmp); 499 #else 500 pushl(Address(tmp, JvmtiThreadState::earlyret_tos_offset())); 501 call_VM_leaf(CAST_FROM_FN_PTR(address, Interpreter::remove_activation_early_entry), 1); 502 #endif // _LP64 503 jmp(rax); 504 bind(L); 505 NOT_LP64(get_thread(java_thread);) 506 } 507 } 508 509 void InterpreterMacroAssembler::get_unsigned_2_byte_index_at_bcp(Register reg, int bcp_offset) { 510 assert(bcp_offset >= 0, "bcp is still pointing to start of bytecode"); 511 load_unsigned_short(reg, Address(_bcp_register, bcp_offset)); 512 bswapl(reg); 513 shrl(reg, 16); 514 } 515 516 void InterpreterMacroAssembler::get_cache_index_at_bcp(Register index, 517 int bcp_offset, 518 size_t index_size) { 519 assert(bcp_offset > 0, "bcp is still pointing to start of bytecode"); 520 if (index_size == sizeof(u2)) { 521 load_unsigned_short(index, Address(_bcp_register, bcp_offset)); 522 } else if (index_size == sizeof(u4)) { 523 movl(index, Address(_bcp_register, bcp_offset)); 524 } else if (index_size == sizeof(u1)) { 525 load_unsigned_byte(index, Address(_bcp_register, bcp_offset)); 526 } else { 527 ShouldNotReachHere(); 528 } 529 } 530 531 // Load object from cpool->resolved_references(index) 532 void InterpreterMacroAssembler::load_resolved_reference_at_index(Register result, 533 Register index, 534 Register tmp) { 535 assert_different_registers(result, index); 536 537 get_constant_pool(result); 538 // load pointer for resolved_references[] objArray 539 movptr(result, Address(result, ConstantPool::cache_offset())); 540 movptr(result, Address(result, ConstantPoolCache::resolved_references_offset())); 541 resolve_oop_handle(result, tmp); 542 load_heap_oop(result, Address(result, index, 543 UseCompressedOops ? Address::times_4 : Address::times_ptr, 544 arrayOopDesc::base_offset_in_bytes(T_OBJECT)), tmp); 545 } 546 547 // load cpool->resolved_klass_at(index) 548 void InterpreterMacroAssembler::load_resolved_klass_at_index(Register klass, 549 Register cpool, 550 Register index) { 551 assert_different_registers(cpool, index); 552 553 movw(index, Address(cpool, index, Address::times_ptr, sizeof(ConstantPool))); 554 Register resolved_klasses = cpool; 555 movptr(resolved_klasses, Address(cpool, ConstantPool::resolved_klasses_offset())); 556 movptr(klass, Address(resolved_klasses, index, Address::times_ptr, Array<Klass*>::base_offset_in_bytes())); 557 } 558 559 // Generate a subtype check: branch to ok_is_subtype if sub_klass is a 560 // subtype of super_klass. 561 // 562 // Args: 563 // rax: superklass 564 // Rsub_klass: subklass 565 // 566 // Kills: 567 // rcx, rdi 568 void InterpreterMacroAssembler::gen_subtype_check(Register Rsub_klass, 569 Label& ok_is_subtype) { 570 assert(Rsub_klass != rax, "rax holds superklass"); 571 LP64_ONLY(assert(Rsub_klass != r14, "r14 holds locals");) 572 LP64_ONLY(assert(Rsub_klass != r13, "r13 holds bcp");) 573 assert(Rsub_klass != rcx, "rcx holds 2ndary super array length"); 574 assert(Rsub_klass != rdi, "rdi holds 2ndary super array scan ptr"); 575 576 // Profile the not-null value's klass. 577 profile_typecheck(rcx, Rsub_klass, rdi); // blows rcx, reloads rdi 578 579 // Do the check. 580 check_klass_subtype(Rsub_klass, rax, rcx, ok_is_subtype); // blows rcx 581 } 582 583 584 #ifndef _LP64 585 void InterpreterMacroAssembler::f2ieee() { 586 if (IEEEPrecision) { 587 fstp_s(Address(rsp, 0)); 588 fld_s(Address(rsp, 0)); 589 } 590 } 591 592 593 void InterpreterMacroAssembler::d2ieee() { 594 if (IEEEPrecision) { 595 fstp_d(Address(rsp, 0)); 596 fld_d(Address(rsp, 0)); 597 } 598 } 599 #endif // _LP64 600 601 // Java Expression Stack 602 603 void InterpreterMacroAssembler::pop_ptr(Register r) { 604 pop(r); 605 } 606 607 void InterpreterMacroAssembler::push_ptr(Register r) { 608 push(r); 609 } 610 611 void InterpreterMacroAssembler::push_i(Register r) { 612 push(r); 613 } 614 615 void InterpreterMacroAssembler::push_i_or_ptr(Register r) { 616 push(r); 617 } 618 619 void InterpreterMacroAssembler::push_f(XMMRegister r) { 620 subptr(rsp, wordSize); 621 movflt(Address(rsp, 0), r); 622 } 623 624 void InterpreterMacroAssembler::pop_f(XMMRegister r) { 625 movflt(r, Address(rsp, 0)); 626 addptr(rsp, wordSize); 627 } 628 629 void InterpreterMacroAssembler::push_d(XMMRegister r) { 630 subptr(rsp, 2 * wordSize); 631 movdbl(Address(rsp, 0), r); 632 } 633 634 void InterpreterMacroAssembler::pop_d(XMMRegister r) { 635 movdbl(r, Address(rsp, 0)); 636 addptr(rsp, 2 * Interpreter::stackElementSize); 637 } 638 639 #ifdef _LP64 640 void InterpreterMacroAssembler::pop_i(Register r) { 641 // XXX can't use pop currently, upper half non clean 642 movl(r, Address(rsp, 0)); 643 addptr(rsp, wordSize); 644 } 645 646 void InterpreterMacroAssembler::pop_l(Register r) { 647 movq(r, Address(rsp, 0)); 648 addptr(rsp, 2 * Interpreter::stackElementSize); 649 } 650 651 void InterpreterMacroAssembler::push_l(Register r) { 652 subptr(rsp, 2 * wordSize); 653 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(0)), r ); 654 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(1)), NULL_WORD ); 655 } 656 657 void InterpreterMacroAssembler::pop(TosState state) { 658 switch (state) { 659 case atos: pop_ptr(); break; 660 case btos: 661 case ztos: 662 case ctos: 663 case stos: 664 case itos: pop_i(); break; 665 case ltos: pop_l(); break; 666 case ftos: pop_f(xmm0); break; 667 case dtos: pop_d(xmm0); break; 668 case vtos: /* nothing to do */ break; 669 default: ShouldNotReachHere(); 670 } 671 interp_verify_oop(rax, state); 672 } 673 674 void InterpreterMacroAssembler::push(TosState state) { 675 interp_verify_oop(rax, state); 676 switch (state) { 677 case atos: push_ptr(); break; 678 case btos: 679 case ztos: 680 case ctos: 681 case stos: 682 case itos: push_i(); break; 683 case ltos: push_l(); break; 684 case ftos: push_f(xmm0); break; 685 case dtos: push_d(xmm0); break; 686 case vtos: /* nothing to do */ break; 687 default : ShouldNotReachHere(); 688 } 689 } 690 #else 691 void InterpreterMacroAssembler::pop_i(Register r) { 692 pop(r); 693 } 694 695 void InterpreterMacroAssembler::pop_l(Register lo, Register hi) { 696 pop(lo); 697 pop(hi); 698 } 699 700 void InterpreterMacroAssembler::pop_f() { 701 fld_s(Address(rsp, 0)); 702 addptr(rsp, 1 * wordSize); 703 } 704 705 void InterpreterMacroAssembler::pop_d() { 706 fld_d(Address(rsp, 0)); 707 addptr(rsp, 2 * wordSize); 708 } 709 710 711 void InterpreterMacroAssembler::pop(TosState state) { 712 switch (state) { 713 case atos: pop_ptr(rax); break; 714 case btos: // fall through 715 case ztos: // fall through 716 case ctos: // fall through 717 case stos: // fall through 718 case itos: pop_i(rax); break; 719 case ltos: pop_l(rax, rdx); break; 720 case ftos: 721 if (UseSSE >= 1) { 722 pop_f(xmm0); 723 } else { 724 pop_f(); 725 } 726 break; 727 case dtos: 728 if (UseSSE >= 2) { 729 pop_d(xmm0); 730 } else { 731 pop_d(); 732 } 733 break; 734 case vtos: /* nothing to do */ break; 735 default : ShouldNotReachHere(); 736 } 737 interp_verify_oop(rax, state); 738 } 739 740 741 void InterpreterMacroAssembler::push_l(Register lo, Register hi) { 742 push(hi); 743 push(lo); 744 } 745 746 void InterpreterMacroAssembler::push_f() { 747 // Do not schedule for no AGI! Never write beyond rsp! 748 subptr(rsp, 1 * wordSize); 749 fstp_s(Address(rsp, 0)); 750 } 751 752 void InterpreterMacroAssembler::push_d() { 753 // Do not schedule for no AGI! Never write beyond rsp! 754 subptr(rsp, 2 * wordSize); 755 fstp_d(Address(rsp, 0)); 756 } 757 758 759 void InterpreterMacroAssembler::push(TosState state) { 760 interp_verify_oop(rax, state); 761 switch (state) { 762 case atos: push_ptr(rax); break; 763 case btos: // fall through 764 case ztos: // fall through 765 case ctos: // fall through 766 case stos: // fall through 767 case itos: push_i(rax); break; 768 case ltos: push_l(rax, rdx); break; 769 case ftos: 770 if (UseSSE >= 1) { 771 push_f(xmm0); 772 } else { 773 push_f(); 774 } 775 break; 776 case dtos: 777 if (UseSSE >= 2) { 778 push_d(xmm0); 779 } else { 780 push_d(); 781 } 782 break; 783 case vtos: /* nothing to do */ break; 784 default : ShouldNotReachHere(); 785 } 786 } 787 #endif // _LP64 788 789 790 // Helpers for swap and dup 791 void InterpreterMacroAssembler::load_ptr(int n, Register val) { 792 movptr(val, Address(rsp, Interpreter::expr_offset_in_bytes(n))); 793 } 794 795 void InterpreterMacroAssembler::store_ptr(int n, Register val) { 796 movptr(Address(rsp, Interpreter::expr_offset_in_bytes(n)), val); 797 } 798 799 800 void InterpreterMacroAssembler::prepare_to_jump_from_interpreted() { 801 // set sender sp 802 lea(_bcp_register, Address(rsp, wordSize)); 803 // record last_sp 804 mov(rcx, _bcp_register); 805 subptr(rcx, rbp); 806 sarptr(rcx, LogBytesPerWord); 807 movptr(Address(rbp, frame::interpreter_frame_last_sp_offset * wordSize), rcx); 808 } 809 810 811 // Jump to from_interpreted entry of a call unless single stepping is possible 812 // in this thread in which case we must call the i2i entry 813 void InterpreterMacroAssembler::jump_from_interpreted(Register method, Register temp) { 814 prepare_to_jump_from_interpreted(); 815 816 if (JvmtiExport::can_post_interpreter_events()) { 817 Label run_compiled_code; 818 // JVMTI events, such as single-stepping, are implemented partly by avoiding running 819 // compiled code in threads for which the event is enabled. Check here for 820 // interp_only_mode if these events CAN be enabled. 821 // interp_only is an int, on little endian it is sufficient to test the byte only 822 // Is a cmpl faster? 823 LP64_ONLY(temp = r15_thread;) 824 NOT_LP64(get_thread(temp);) 825 cmpb(Address(temp, JavaThread::interp_only_mode_offset()), 0); 826 jccb(Assembler::zero, run_compiled_code); 827 jmp(Address(method, Method::interpreter_entry_offset())); 828 bind(run_compiled_code); 829 } 830 831 jmp(Address(method, Method::from_interpreted_offset())); 832 } 833 834 // The following two routines provide a hook so that an implementation 835 // can schedule the dispatch in two parts. x86 does not do this. 836 void InterpreterMacroAssembler::dispatch_prolog(TosState state, int step) { 837 // Nothing x86 specific to be done here 838 } 839 840 void InterpreterMacroAssembler::dispatch_epilog(TosState state, int step) { 841 dispatch_next(state, step); 842 } 843 844 void InterpreterMacroAssembler::dispatch_base(TosState state, 845 address* table, 846 bool verifyoop, 847 bool generate_poll) { 848 verify_FPU(1, state); 849 if (VerifyActivationFrameSize) { 850 Label L; 851 mov(rcx, rbp); 852 subptr(rcx, rsp); 853 int32_t min_frame_size = 854 (frame::link_offset - frame::interpreter_frame_initial_sp_offset) * 855 wordSize; 856 cmpptr(rcx, min_frame_size); 857 jcc(Assembler::greaterEqual, L); 858 stop("broken stack frame"); 859 bind(L); 860 } 861 if (verifyoop) { 862 interp_verify_oop(rax, state); 863 } 864 865 address* const safepoint_table = Interpreter::safept_table(state); 866 #ifdef _LP64 867 Label no_safepoint, dispatch; 868 if (table != safepoint_table && generate_poll) { 869 NOT_PRODUCT(block_comment("Thread-local Safepoint poll")); 870 testb(Address(r15_thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit()); 871 872 jccb(Assembler::zero, no_safepoint); 873 lea(rscratch1, ExternalAddress((address)safepoint_table)); 874 jmpb(dispatch); 875 } 876 877 bind(no_safepoint); 878 lea(rscratch1, ExternalAddress((address)table)); 879 bind(dispatch); 880 jmp(Address(rscratch1, rbx, Address::times_8)); 881 882 #else 883 Address index(noreg, rbx, Address::times_ptr); 884 if (table != safepoint_table && generate_poll) { 885 NOT_PRODUCT(block_comment("Thread-local Safepoint poll")); 886 Label no_safepoint; 887 const Register thread = rcx; 888 get_thread(thread); 889 testb(Address(thread, JavaThread::polling_word_offset()), SafepointMechanism::poll_bit()); 890 891 jccb(Assembler::zero, no_safepoint); 892 ArrayAddress dispatch_addr(ExternalAddress((address)safepoint_table), index); 893 jump(dispatch_addr, noreg); 894 bind(no_safepoint); 895 } 896 897 { 898 ArrayAddress dispatch_addr(ExternalAddress((address)table), index); 899 jump(dispatch_addr, noreg); 900 } 901 #endif // _LP64 902 } 903 904 void InterpreterMacroAssembler::dispatch_only(TosState state, bool generate_poll) { 905 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll); 906 } 907 908 void InterpreterMacroAssembler::dispatch_only_normal(TosState state) { 909 dispatch_base(state, Interpreter::normal_table(state)); 910 } 911 912 void InterpreterMacroAssembler::dispatch_only_noverify(TosState state) { 913 dispatch_base(state, Interpreter::normal_table(state), false); 914 } 915 916 917 void InterpreterMacroAssembler::dispatch_next(TosState state, int step, bool generate_poll) { 918 // load next bytecode (load before advancing _bcp_register to prevent AGI) 919 load_unsigned_byte(rbx, Address(_bcp_register, step)); 920 // advance _bcp_register 921 increment(_bcp_register, step); 922 dispatch_base(state, Interpreter::dispatch_table(state), true, generate_poll); 923 } 924 925 void InterpreterMacroAssembler::dispatch_via(TosState state, address* table) { 926 // load current bytecode 927 load_unsigned_byte(rbx, Address(_bcp_register, 0)); 928 dispatch_base(state, table); 929 } 930 931 void InterpreterMacroAssembler::narrow(Register result) { 932 933 // Get method->_constMethod->_result_type 934 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize)); 935 movptr(rcx, Address(rcx, Method::const_offset())); 936 load_unsigned_byte(rcx, Address(rcx, ConstMethod::result_type_offset())); 937 938 Label done, notBool, notByte, notChar; 939 940 // common case first 941 cmpl(rcx, T_INT); 942 jcc(Assembler::equal, done); 943 944 // mask integer result to narrower return type. 945 cmpl(rcx, T_BOOLEAN); 946 jcc(Assembler::notEqual, notBool); 947 andl(result, 0x1); 948 jmp(done); 949 950 bind(notBool); 951 cmpl(rcx, T_BYTE); 952 jcc(Assembler::notEqual, notByte); 953 LP64_ONLY(movsbl(result, result);) 954 NOT_LP64(shll(result, 24);) // truncate upper 24 bits 955 NOT_LP64(sarl(result, 24);) // and sign-extend byte 956 jmp(done); 957 958 bind(notByte); 959 cmpl(rcx, T_CHAR); 960 jcc(Assembler::notEqual, notChar); 961 LP64_ONLY(movzwl(result, result);) 962 NOT_LP64(andl(result, 0xFFFF);) // truncate upper 16 bits 963 jmp(done); 964 965 bind(notChar); 966 // cmpl(rcx, T_SHORT); // all that's left 967 // jcc(Assembler::notEqual, done); 968 LP64_ONLY(movswl(result, result);) 969 NOT_LP64(shll(result, 16);) // truncate upper 16 bits 970 NOT_LP64(sarl(result, 16);) // and sign-extend short 971 972 // Nothing to do for T_INT 973 bind(done); 974 } 975 976 // remove activation 977 // 978 // Apply stack watermark barrier. 979 // Unlock the receiver if this is a synchronized method. 980 // Unlock any Java monitors from synchronized blocks. 981 // Remove the activation from the stack. 982 // 983 // If there are locked Java monitors 984 // If throw_monitor_exception 985 // throws IllegalMonitorStateException 986 // Else if install_monitor_exception 987 // installs IllegalMonitorStateException 988 // Else 989 // no error processing 990 void InterpreterMacroAssembler::remove_activation( 991 TosState state, 992 Register ret_addr, 993 bool throw_monitor_exception, 994 bool install_monitor_exception, 995 bool notify_jvmdi) { 996 // Note: Registers rdx xmm0 may be in use for the 997 // result check if synchronized method 998 Label unlocked, unlock, no_unlock; 999 1000 const Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx); 1001 const Register robj = LP64_ONLY(c_rarg1) NOT_LP64(rdx); 1002 const Register rmon = LP64_ONLY(c_rarg1) NOT_LP64(rcx); 1003 // monitor pointers need different register 1004 // because rdx may have the result in it 1005 NOT_LP64(get_thread(rthread);) 1006 1007 // The below poll is for the stack watermark barrier. It allows fixing up frames lazily, 1008 // that would normally not be safe to use. Such bad returns into unsafe territory of 1009 // the stack, will call InterpreterRuntime::at_unwind. 1010 Label slow_path; 1011 Label fast_path; 1012 safepoint_poll(slow_path, rthread, true /* at_return */, false /* in_nmethod */); 1013 jmp(fast_path); 1014 bind(slow_path); 1015 push(state); 1016 set_last_Java_frame(rthread, noreg, rbp, (address)pc(), rscratch1); 1017 super_call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::at_unwind), rthread); 1018 NOT_LP64(get_thread(rthread);) // call_VM clobbered it, restore 1019 reset_last_Java_frame(rthread, true); 1020 pop(state); 1021 bind(fast_path); 1022 1023 // get the value of _do_not_unlock_if_synchronized into rdx 1024 const Address do_not_unlock_if_synchronized(rthread, 1025 in_bytes(JavaThread::do_not_unlock_if_synchronized_offset())); 1026 movbool(rbx, do_not_unlock_if_synchronized); 1027 movbool(do_not_unlock_if_synchronized, false); // reset the flag 1028 1029 // get method access flags 1030 movptr(rcx, Address(rbp, frame::interpreter_frame_method_offset * wordSize)); 1031 movl(rcx, Address(rcx, Method::access_flags_offset())); 1032 testl(rcx, JVM_ACC_SYNCHRONIZED); 1033 jcc(Assembler::zero, unlocked); 1034 1035 // Don't unlock anything if the _do_not_unlock_if_synchronized flag 1036 // is set. 1037 testbool(rbx); 1038 jcc(Assembler::notZero, no_unlock); 1039 1040 // unlock monitor 1041 push(state); // save result 1042 1043 // BasicObjectLock will be first in list, since this is a 1044 // synchronized method. However, need to check that the object has 1045 // not been unlocked by an explicit monitorexit bytecode. 1046 const Address monitor(rbp, frame::interpreter_frame_initial_sp_offset * 1047 wordSize - (int) sizeof(BasicObjectLock)); 1048 // We use c_rarg1/rdx so that if we go slow path it will be the correct 1049 // register for unlock_object to pass to VM directly 1050 lea(robj, monitor); // address of first monitor 1051 1052 movptr(rax, Address(robj, BasicObjectLock::obj_offset())); 1053 testptr(rax, rax); 1054 jcc(Assembler::notZero, unlock); 1055 1056 pop(state); 1057 if (throw_monitor_exception) { 1058 // Entry already unlocked, need to throw exception 1059 NOT_LP64(empty_FPU_stack();) // remove possible return value from FPU-stack, otherwise stack could overflow 1060 call_VM(noreg, CAST_FROM_FN_PTR(address, 1061 InterpreterRuntime::throw_illegal_monitor_state_exception)); 1062 should_not_reach_here(); 1063 } else { 1064 // Monitor already unlocked during a stack unroll. If requested, 1065 // install an illegal_monitor_state_exception. Continue with 1066 // stack unrolling. 1067 if (install_monitor_exception) { 1068 NOT_LP64(empty_FPU_stack();) 1069 call_VM(noreg, CAST_FROM_FN_PTR(address, 1070 InterpreterRuntime::new_illegal_monitor_state_exception)); 1071 } 1072 jmp(unlocked); 1073 } 1074 1075 bind(unlock); 1076 unlock_object(robj); 1077 pop(state); 1078 1079 // Check that for block-structured locking (i.e., that all locked 1080 // objects has been unlocked) 1081 bind(unlocked); 1082 1083 // rax, rdx: Might contain return value 1084 1085 // Check that all monitors are unlocked 1086 { 1087 Label loop, exception, entry, restart; 1088 const int entry_size = frame::interpreter_frame_monitor_size_in_bytes(); 1089 const Address monitor_block_top( 1090 rbp, frame::interpreter_frame_monitor_block_top_offset * wordSize); 1091 const Address monitor_block_bot( 1092 rbp, frame::interpreter_frame_initial_sp_offset * wordSize); 1093 1094 bind(restart); 1095 // We use c_rarg1 so that if we go slow path it will be the correct 1096 // register for unlock_object to pass to VM directly 1097 movptr(rmon, monitor_block_top); // derelativize pointer 1098 lea(rmon, Address(rbp, rmon, Address::times_ptr)); 1099 // c_rarg1 points to current entry, starting with top-most entry 1100 1101 lea(rbx, monitor_block_bot); // points to word before bottom of 1102 // monitor block 1103 jmp(entry); 1104 1105 // Entry already locked, need to throw exception 1106 bind(exception); 1107 1108 if (throw_monitor_exception) { 1109 // Throw exception 1110 NOT_LP64(empty_FPU_stack();) 1111 MacroAssembler::call_VM(noreg, 1112 CAST_FROM_FN_PTR(address, InterpreterRuntime:: 1113 throw_illegal_monitor_state_exception)); 1114 should_not_reach_here(); 1115 } else { 1116 // Stack unrolling. Unlock object and install illegal_monitor_exception. 1117 // Unlock does not block, so don't have to worry about the frame. 1118 // We don't have to preserve c_rarg1 since we are going to throw an exception. 1119 1120 push(state); 1121 mov(robj, rmon); // nop if robj and rmon are the same 1122 unlock_object(robj); 1123 pop(state); 1124 1125 if (install_monitor_exception) { 1126 NOT_LP64(empty_FPU_stack();) 1127 call_VM(noreg, CAST_FROM_FN_PTR(address, 1128 InterpreterRuntime:: 1129 new_illegal_monitor_state_exception)); 1130 } 1131 1132 jmp(restart); 1133 } 1134 1135 bind(loop); 1136 // check if current entry is used 1137 cmpptr(Address(rmon, BasicObjectLock::obj_offset()), NULL_WORD); 1138 jcc(Assembler::notEqual, exception); 1139 1140 addptr(rmon, entry_size); // otherwise advance to next entry 1141 bind(entry); 1142 cmpptr(rmon, rbx); // check if bottom reached 1143 jcc(Assembler::notEqual, loop); // if not at bottom then check this entry 1144 } 1145 1146 bind(no_unlock); 1147 1148 // jvmti support 1149 if (notify_jvmdi) { 1150 notify_method_exit(state, NotifyJVMTI); // preserve TOSCA 1151 } else { 1152 notify_method_exit(state, SkipNotifyJVMTI); // preserve TOSCA 1153 } 1154 1155 // remove activation 1156 // get sender sp 1157 movptr(rbx, 1158 Address(rbp, frame::interpreter_frame_sender_sp_offset * wordSize)); 1159 if (StackReservedPages > 0) { 1160 // testing if reserved zone needs to be re-enabled 1161 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx); 1162 Label no_reserved_zone_enabling; 1163 1164 NOT_LP64(get_thread(rthread);) 1165 1166 // check if already enabled - if so no re-enabling needed 1167 assert(sizeof(StackOverflow::StackGuardState) == 4, "unexpected size"); 1168 cmpl(Address(rthread, JavaThread::stack_guard_state_offset()), StackOverflow::stack_guard_enabled); 1169 jcc(Assembler::equal, no_reserved_zone_enabling); 1170 1171 cmpptr(rbx, Address(rthread, JavaThread::reserved_stack_activation_offset())); 1172 jcc(Assembler::lessEqual, no_reserved_zone_enabling); 1173 1174 call_VM_leaf( 1175 CAST_FROM_FN_PTR(address, SharedRuntime::enable_stack_reserved_zone), rthread); 1176 call_VM(noreg, CAST_FROM_FN_PTR(address, 1177 InterpreterRuntime::throw_delayed_StackOverflowError)); 1178 should_not_reach_here(); 1179 1180 bind(no_reserved_zone_enabling); 1181 } 1182 leave(); // remove frame anchor 1183 pop(ret_addr); // get return address 1184 mov(rsp, rbx); // set sp to sender sp 1185 pop_cont_fastpath(); 1186 } 1187 1188 void InterpreterMacroAssembler::get_method_counters(Register method, 1189 Register mcs, Label& skip) { 1190 Label has_counters; 1191 movptr(mcs, Address(method, Method::method_counters_offset())); 1192 testptr(mcs, mcs); 1193 jcc(Assembler::notZero, has_counters); 1194 call_VM(noreg, CAST_FROM_FN_PTR(address, 1195 InterpreterRuntime::build_method_counters), method); 1196 movptr(mcs, Address(method,Method::method_counters_offset())); 1197 testptr(mcs, mcs); 1198 jcc(Assembler::zero, skip); // No MethodCounters allocated, OutOfMemory 1199 bind(has_counters); 1200 } 1201 1202 1203 // Lock object 1204 // 1205 // Args: 1206 // rdx, c_rarg1: BasicObjectLock to be used for locking 1207 // 1208 // Kills: 1209 // rax, rbx 1210 void InterpreterMacroAssembler::lock_object(Register lock_reg) { 1211 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx), 1212 "The argument is only for looks. It must be c_rarg1"); 1213 1214 if (LockingMode == LM_MONITOR) { 1215 call_VM_preemptable(noreg, 1216 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 1217 lock_reg); 1218 } else { 1219 Label count_locking, done, slow_case; 1220 1221 const Register swap_reg = rax; // Must use rax for cmpxchg instruction 1222 const Register tmp_reg = rbx; 1223 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop 1224 const Register rklass_decode_tmp = rscratch1; 1225 1226 const int obj_offset = in_bytes(BasicObjectLock::obj_offset()); 1227 const int lock_offset = in_bytes(BasicObjectLock::lock_offset()); 1228 const int mark_offset = lock_offset + 1229 BasicLock::displaced_header_offset_in_bytes(); 1230 1231 // Load object pointer into obj_reg 1232 movptr(obj_reg, Address(lock_reg, obj_offset)); 1233 1234 if (DiagnoseSyncOnValueBasedClasses != 0) { 1235 load_klass(tmp_reg, obj_reg, rklass_decode_tmp); 1236 testb(Address(tmp_reg, Klass::misc_flags_offset()), KlassFlags::_misc_is_value_based_class); 1237 jcc(Assembler::notZero, slow_case); 1238 } 1239 1240 if (LockingMode == LM_LIGHTWEIGHT) { 1241 #ifdef _LP64 1242 const Register thread = r15_thread; 1243 lightweight_lock(lock_reg, obj_reg, swap_reg, thread, tmp_reg, slow_case); 1244 #else 1245 // Lacking registers and thread on x86_32. Always take slow path. 1246 jmp(slow_case); 1247 #endif 1248 } else if (LockingMode == LM_LEGACY) { 1249 // Load immediate 1 into swap_reg %rax 1250 movl(swap_reg, 1); 1251 1252 // Load (object->mark() | 1) into swap_reg %rax 1253 orptr(swap_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes())); 1254 1255 // Save (object->mark() | 1) into BasicLock's displaced header 1256 movptr(Address(lock_reg, mark_offset), swap_reg); 1257 1258 assert(lock_offset == 0, 1259 "displaced header must be first word in BasicObjectLock"); 1260 1261 lock(); 1262 cmpxchgptr(lock_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes())); 1263 jcc(Assembler::zero, count_locking); 1264 1265 const int zero_bits = LP64_ONLY(7) NOT_LP64(3); 1266 1267 // Fast check for recursive lock. 1268 // 1269 // Can apply the optimization only if this is a stack lock 1270 // allocated in this thread. For efficiency, we can focus on 1271 // recently allocated stack locks (instead of reading the stack 1272 // base and checking whether 'mark' points inside the current 1273 // thread stack): 1274 // 1) (mark & zero_bits) == 0, and 1275 // 2) rsp <= mark < mark + os::pagesize() 1276 // 1277 // Warning: rsp + os::pagesize can overflow the stack base. We must 1278 // neither apply the optimization for an inflated lock allocated 1279 // just above the thread stack (this is why condition 1 matters) 1280 // nor apply the optimization if the stack lock is inside the stack 1281 // of another thread. The latter is avoided even in case of overflow 1282 // because we have guard pages at the end of all stacks. Hence, if 1283 // we go over the stack base and hit the stack of another thread, 1284 // this should not be in a writeable area that could contain a 1285 // stack lock allocated by that thread. As a consequence, a stack 1286 // lock less than page size away from rsp is guaranteed to be 1287 // owned by the current thread. 1288 // 1289 // These 3 tests can be done by evaluating the following 1290 // expression: ((mark - rsp) & (zero_bits - os::vm_page_size())), 1291 // assuming both stack pointer and pagesize have their 1292 // least significant bits clear. 1293 // NOTE: the mark is in swap_reg %rax as the result of cmpxchg 1294 subptr(swap_reg, rsp); 1295 andptr(swap_reg, zero_bits - (int)os::vm_page_size()); 1296 1297 // Save the test result, for recursive case, the result is zero 1298 movptr(Address(lock_reg, mark_offset), swap_reg); 1299 jcc(Assembler::notZero, slow_case); 1300 1301 bind(count_locking); 1302 inc_held_monitor_count(); 1303 } 1304 jmp(done); 1305 1306 bind(slow_case); 1307 1308 // Call the runtime routine for slow case 1309 call_VM_preemptable(noreg, 1310 CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorenter), 1311 lock_reg); 1312 bind(done); 1313 } 1314 } 1315 1316 1317 // Unlocks an object. Used in monitorexit bytecode and 1318 // remove_activation. Throws an IllegalMonitorException if object is 1319 // not locked by current thread. 1320 // 1321 // Args: 1322 // rdx, c_rarg1: BasicObjectLock for lock 1323 // 1324 // Kills: 1325 // rax 1326 // c_rarg0, c_rarg1, c_rarg2, c_rarg3, ... (param regs) 1327 // rscratch1 (scratch reg) 1328 // rax, rbx, rcx, rdx 1329 void InterpreterMacroAssembler::unlock_object(Register lock_reg) { 1330 assert(lock_reg == LP64_ONLY(c_rarg1) NOT_LP64(rdx), 1331 "The argument is only for looks. It must be c_rarg1"); 1332 1333 if (LockingMode == LM_MONITOR) { 1334 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg); 1335 } else { 1336 Label count_locking, done, slow_case; 1337 1338 const Register swap_reg = rax; // Must use rax for cmpxchg instruction 1339 const Register header_reg = LP64_ONLY(c_rarg2) NOT_LP64(rbx); // Will contain the old oopMark 1340 const Register obj_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); // Will contain the oop 1341 1342 save_bcp(); // Save in case of exception 1343 1344 if (LockingMode != LM_LIGHTWEIGHT) { 1345 // Convert from BasicObjectLock structure to object and BasicLock 1346 // structure Store the BasicLock address into %rax 1347 lea(swap_reg, Address(lock_reg, BasicObjectLock::lock_offset())); 1348 } 1349 1350 // Load oop into obj_reg(%c_rarg3) 1351 movptr(obj_reg, Address(lock_reg, BasicObjectLock::obj_offset())); 1352 1353 // Free entry 1354 movptr(Address(lock_reg, BasicObjectLock::obj_offset()), NULL_WORD); 1355 1356 if (LockingMode == LM_LIGHTWEIGHT) { 1357 #ifdef _LP64 1358 lightweight_unlock(obj_reg, swap_reg, r15_thread, header_reg, slow_case); 1359 #else 1360 // Lacking registers and thread on x86_32. Always take slow path. 1361 jmp(slow_case); 1362 #endif 1363 } else if (LockingMode == LM_LEGACY) { 1364 // Load the old header from BasicLock structure 1365 movptr(header_reg, Address(swap_reg, 1366 BasicLock::displaced_header_offset_in_bytes())); 1367 1368 // Test for recursion 1369 testptr(header_reg, header_reg); 1370 1371 // zero for recursive case 1372 jcc(Assembler::zero, count_locking); 1373 1374 // Atomic swap back the old header 1375 lock(); 1376 cmpxchgptr(header_reg, Address(obj_reg, oopDesc::mark_offset_in_bytes())); 1377 1378 // zero for simple unlock of a stack-lock case 1379 jcc(Assembler::notZero, slow_case); 1380 1381 bind(count_locking); 1382 dec_held_monitor_count(); 1383 } 1384 jmp(done); 1385 1386 bind(slow_case); 1387 // Call the runtime routine for slow case. 1388 movptr(Address(lock_reg, BasicObjectLock::obj_offset()), obj_reg); // restore obj 1389 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::monitorexit), lock_reg); 1390 1391 bind(done); 1392 1393 restore_bcp(); 1394 } 1395 } 1396 1397 void InterpreterMacroAssembler::test_method_data_pointer(Register mdp, 1398 Label& zero_continue) { 1399 assert(ProfileInterpreter, "must be profiling interpreter"); 1400 movptr(mdp, Address(rbp, frame::interpreter_frame_mdp_offset * wordSize)); 1401 testptr(mdp, mdp); 1402 jcc(Assembler::zero, zero_continue); 1403 } 1404 1405 1406 // Set the method data pointer for the current bcp. 1407 void InterpreterMacroAssembler::set_method_data_pointer_for_bcp() { 1408 assert(ProfileInterpreter, "must be profiling interpreter"); 1409 Label set_mdp; 1410 push(rax); 1411 push(rbx); 1412 1413 get_method(rbx); 1414 // Test MDO to avoid the call if it is null. 1415 movptr(rax, Address(rbx, in_bytes(Method::method_data_offset()))); 1416 testptr(rax, rax); 1417 jcc(Assembler::zero, set_mdp); 1418 // rbx: method 1419 // _bcp_register: bcp 1420 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::bcp_to_di), rbx, _bcp_register); 1421 // rax: mdi 1422 // mdo is guaranteed to be non-zero here, we checked for it before the call. 1423 movptr(rbx, Address(rbx, in_bytes(Method::method_data_offset()))); 1424 addptr(rbx, in_bytes(MethodData::data_offset())); 1425 addptr(rax, rbx); 1426 bind(set_mdp); 1427 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), rax); 1428 pop(rbx); 1429 pop(rax); 1430 } 1431 1432 void InterpreterMacroAssembler::verify_method_data_pointer() { 1433 assert(ProfileInterpreter, "must be profiling interpreter"); 1434 #ifdef ASSERT 1435 Label verify_continue; 1436 push(rax); 1437 push(rbx); 1438 Register arg3_reg = LP64_ONLY(c_rarg3) NOT_LP64(rcx); 1439 Register arg2_reg = LP64_ONLY(c_rarg2) NOT_LP64(rdx); 1440 push(arg3_reg); 1441 push(arg2_reg); 1442 test_method_data_pointer(arg3_reg, verify_continue); // If mdp is zero, continue 1443 get_method(rbx); 1444 1445 // If the mdp is valid, it will point to a DataLayout header which is 1446 // consistent with the bcp. The converse is highly probable also. 1447 load_unsigned_short(arg2_reg, 1448 Address(arg3_reg, in_bytes(DataLayout::bci_offset()))); 1449 addptr(arg2_reg, Address(rbx, Method::const_offset())); 1450 lea(arg2_reg, Address(arg2_reg, ConstMethod::codes_offset())); 1451 cmpptr(arg2_reg, _bcp_register); 1452 jcc(Assembler::equal, verify_continue); 1453 // rbx: method 1454 // _bcp_register: bcp 1455 // c_rarg3: mdp 1456 call_VM_leaf(CAST_FROM_FN_PTR(address, InterpreterRuntime::verify_mdp), 1457 rbx, _bcp_register, arg3_reg); 1458 bind(verify_continue); 1459 pop(arg2_reg); 1460 pop(arg3_reg); 1461 pop(rbx); 1462 pop(rax); 1463 #endif // ASSERT 1464 } 1465 1466 1467 void InterpreterMacroAssembler::set_mdp_data_at(Register mdp_in, 1468 int constant, 1469 Register value) { 1470 assert(ProfileInterpreter, "must be profiling interpreter"); 1471 Address data(mdp_in, constant); 1472 movptr(data, value); 1473 } 1474 1475 1476 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in, 1477 int constant, 1478 bool decrement) { 1479 // Counter address 1480 Address data(mdp_in, constant); 1481 1482 increment_mdp_data_at(data, decrement); 1483 } 1484 1485 void InterpreterMacroAssembler::increment_mdp_data_at(Address data, 1486 bool decrement) { 1487 assert(ProfileInterpreter, "must be profiling interpreter"); 1488 // %%% this does 64bit counters at best it is wasting space 1489 // at worst it is a rare bug when counters overflow 1490 1491 if (decrement) { 1492 // Decrement the register. Set condition codes. 1493 addptr(data, -DataLayout::counter_increment); 1494 // If the decrement causes the counter to overflow, stay negative 1495 Label L; 1496 jcc(Assembler::negative, L); 1497 addptr(data, DataLayout::counter_increment); 1498 bind(L); 1499 } else { 1500 assert(DataLayout::counter_increment == 1, 1501 "flow-free idiom only works with 1"); 1502 // Increment the register. Set carry flag. 1503 addptr(data, DataLayout::counter_increment); 1504 // If the increment causes the counter to overflow, pull back by 1. 1505 sbbptr(data, 0); 1506 } 1507 } 1508 1509 1510 void InterpreterMacroAssembler::increment_mdp_data_at(Register mdp_in, 1511 Register reg, 1512 int constant, 1513 bool decrement) { 1514 Address data(mdp_in, reg, Address::times_1, constant); 1515 1516 increment_mdp_data_at(data, decrement); 1517 } 1518 1519 void InterpreterMacroAssembler::set_mdp_flag_at(Register mdp_in, 1520 int flag_byte_constant) { 1521 assert(ProfileInterpreter, "must be profiling interpreter"); 1522 int header_offset = in_bytes(DataLayout::flags_offset()); 1523 int header_bits = flag_byte_constant; 1524 // Set the flag 1525 orb(Address(mdp_in, header_offset), header_bits); 1526 } 1527 1528 1529 1530 void InterpreterMacroAssembler::test_mdp_data_at(Register mdp_in, 1531 int offset, 1532 Register value, 1533 Register test_value_out, 1534 Label& not_equal_continue) { 1535 assert(ProfileInterpreter, "must be profiling interpreter"); 1536 if (test_value_out == noreg) { 1537 cmpptr(value, Address(mdp_in, offset)); 1538 } else { 1539 // Put the test value into a register, so caller can use it: 1540 movptr(test_value_out, Address(mdp_in, offset)); 1541 cmpptr(test_value_out, value); 1542 } 1543 jcc(Assembler::notEqual, not_equal_continue); 1544 } 1545 1546 1547 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, 1548 int offset_of_disp) { 1549 assert(ProfileInterpreter, "must be profiling interpreter"); 1550 Address disp_address(mdp_in, offset_of_disp); 1551 addptr(mdp_in, disp_address); 1552 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in); 1553 } 1554 1555 1556 void InterpreterMacroAssembler::update_mdp_by_offset(Register mdp_in, 1557 Register reg, 1558 int offset_of_disp) { 1559 assert(ProfileInterpreter, "must be profiling interpreter"); 1560 Address disp_address(mdp_in, reg, Address::times_1, offset_of_disp); 1561 addptr(mdp_in, disp_address); 1562 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in); 1563 } 1564 1565 1566 void InterpreterMacroAssembler::update_mdp_by_constant(Register mdp_in, 1567 int constant) { 1568 assert(ProfileInterpreter, "must be profiling interpreter"); 1569 addptr(mdp_in, constant); 1570 movptr(Address(rbp, frame::interpreter_frame_mdp_offset * wordSize), mdp_in); 1571 } 1572 1573 1574 void InterpreterMacroAssembler::update_mdp_for_ret(Register return_bci) { 1575 assert(ProfileInterpreter, "must be profiling interpreter"); 1576 push(return_bci); // save/restore across call_VM 1577 call_VM(noreg, 1578 CAST_FROM_FN_PTR(address, InterpreterRuntime::update_mdp_for_ret), 1579 return_bci); 1580 pop(return_bci); 1581 } 1582 1583 1584 void InterpreterMacroAssembler::profile_taken_branch(Register mdp, 1585 Register bumped_count) { 1586 if (ProfileInterpreter) { 1587 Label profile_continue; 1588 1589 // If no method data exists, go to profile_continue. 1590 // Otherwise, assign to mdp 1591 test_method_data_pointer(mdp, profile_continue); 1592 1593 // We are taking a branch. Increment the taken count. 1594 // We inline increment_mdp_data_at to return bumped_count in a register 1595 //increment_mdp_data_at(mdp, in_bytes(JumpData::taken_offset())); 1596 Address data(mdp, in_bytes(JumpData::taken_offset())); 1597 movptr(bumped_count, data); 1598 assert(DataLayout::counter_increment == 1, 1599 "flow-free idiom only works with 1"); 1600 addptr(bumped_count, DataLayout::counter_increment); 1601 sbbptr(bumped_count, 0); 1602 movptr(data, bumped_count); // Store back out 1603 1604 // The method data pointer needs to be updated to reflect the new target. 1605 update_mdp_by_offset(mdp, in_bytes(JumpData::displacement_offset())); 1606 bind(profile_continue); 1607 } 1608 } 1609 1610 1611 void InterpreterMacroAssembler::profile_not_taken_branch(Register mdp) { 1612 if (ProfileInterpreter) { 1613 Label profile_continue; 1614 1615 // If no method data exists, go to profile_continue. 1616 test_method_data_pointer(mdp, profile_continue); 1617 1618 // We are taking a branch. Increment the not taken count. 1619 increment_mdp_data_at(mdp, in_bytes(BranchData::not_taken_offset())); 1620 1621 // The method data pointer needs to be updated to correspond to 1622 // the next bytecode 1623 update_mdp_by_constant(mdp, in_bytes(BranchData::branch_data_size())); 1624 bind(profile_continue); 1625 } 1626 } 1627 1628 void InterpreterMacroAssembler::profile_call(Register mdp) { 1629 if (ProfileInterpreter) { 1630 Label profile_continue; 1631 1632 // If no method data exists, go to profile_continue. 1633 test_method_data_pointer(mdp, profile_continue); 1634 1635 // We are making a call. Increment the count. 1636 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1637 1638 // The method data pointer needs to be updated to reflect the new target. 1639 update_mdp_by_constant(mdp, in_bytes(CounterData::counter_data_size())); 1640 bind(profile_continue); 1641 } 1642 } 1643 1644 1645 void InterpreterMacroAssembler::profile_final_call(Register mdp) { 1646 if (ProfileInterpreter) { 1647 Label profile_continue; 1648 1649 // If no method data exists, go to profile_continue. 1650 test_method_data_pointer(mdp, profile_continue); 1651 1652 // We are making a call. Increment the count. 1653 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1654 1655 // The method data pointer needs to be updated to reflect the new target. 1656 update_mdp_by_constant(mdp, 1657 in_bytes(VirtualCallData:: 1658 virtual_call_data_size())); 1659 bind(profile_continue); 1660 } 1661 } 1662 1663 1664 void InterpreterMacroAssembler::profile_virtual_call(Register receiver, 1665 Register mdp, 1666 Register reg2, 1667 bool receiver_can_be_null) { 1668 if (ProfileInterpreter) { 1669 Label profile_continue; 1670 1671 // If no method data exists, go to profile_continue. 1672 test_method_data_pointer(mdp, profile_continue); 1673 1674 Label skip_receiver_profile; 1675 if (receiver_can_be_null) { 1676 Label not_null; 1677 testptr(receiver, receiver); 1678 jccb(Assembler::notZero, not_null); 1679 // We are making a call. Increment the count for null receiver. 1680 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1681 jmp(skip_receiver_profile); 1682 bind(not_null); 1683 } 1684 1685 // Record the receiver type. 1686 record_klass_in_profile(receiver, mdp, reg2, true); 1687 bind(skip_receiver_profile); 1688 1689 // The method data pointer needs to be updated to reflect the new target. 1690 update_mdp_by_constant(mdp, in_bytes(VirtualCallData::virtual_call_data_size())); 1691 bind(profile_continue); 1692 } 1693 } 1694 1695 // This routine creates a state machine for updating the multi-row 1696 // type profile at a virtual call site (or other type-sensitive bytecode). 1697 // The machine visits each row (of receiver/count) until the receiver type 1698 // is found, or until it runs out of rows. At the same time, it remembers 1699 // the location of the first empty row. (An empty row records null for its 1700 // receiver, and can be allocated for a newly-observed receiver type.) 1701 // Because there are two degrees of freedom in the state, a simple linear 1702 // search will not work; it must be a decision tree. Hence this helper 1703 // function is recursive, to generate the required tree structured code. 1704 // It's the interpreter, so we are trading off code space for speed. 1705 // See below for example code. 1706 void InterpreterMacroAssembler::record_klass_in_profile_helper( 1707 Register receiver, Register mdp, 1708 Register reg2, int start_row, 1709 Label& done, bool is_virtual_call) { 1710 if (TypeProfileWidth == 0) { 1711 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1712 } else { 1713 record_item_in_profile_helper(receiver, mdp, reg2, 0, done, TypeProfileWidth, 1714 &VirtualCallData::receiver_offset, &VirtualCallData::receiver_count_offset); 1715 } 1716 } 1717 1718 void InterpreterMacroAssembler::record_item_in_profile_helper(Register item, Register mdp, Register reg2, int start_row, 1719 Label& done, int total_rows, 1720 OffsetFunction item_offset_fn, 1721 OffsetFunction item_count_offset_fn) { 1722 int last_row = total_rows - 1; 1723 assert(start_row <= last_row, "must be work left to do"); 1724 // Test this row for both the item and for null. 1725 // Take any of three different outcomes: 1726 // 1. found item => increment count and goto done 1727 // 2. found null => keep looking for case 1, maybe allocate this cell 1728 // 3. found something else => keep looking for cases 1 and 2 1729 // Case 3 is handled by a recursive call. 1730 for (int row = start_row; row <= last_row; row++) { 1731 Label next_test; 1732 bool test_for_null_also = (row == start_row); 1733 1734 // See if the item is item[n]. 1735 int item_offset = in_bytes(item_offset_fn(row)); 1736 test_mdp_data_at(mdp, item_offset, item, 1737 (test_for_null_also ? reg2 : noreg), 1738 next_test); 1739 // (Reg2 now contains the item from the CallData.) 1740 1741 // The item is item[n]. Increment count[n]. 1742 int count_offset = in_bytes(item_count_offset_fn(row)); 1743 increment_mdp_data_at(mdp, count_offset); 1744 jmp(done); 1745 bind(next_test); 1746 1747 if (test_for_null_also) { 1748 // Failed the equality check on item[n]... Test for null. 1749 testptr(reg2, reg2); 1750 if (start_row == last_row) { 1751 // The only thing left to do is handle the null case. 1752 Label found_null; 1753 jccb(Assembler::zero, found_null); 1754 // Item did not match any saved item and there is no empty row for it. 1755 // Increment total counter to indicate polymorphic case. 1756 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1757 jmp(done); 1758 bind(found_null); 1759 break; 1760 } 1761 Label found_null; 1762 // Since null is rare, make it be the branch-taken case. 1763 jcc(Assembler::zero, found_null); 1764 1765 // Put all the "Case 3" tests here. 1766 record_item_in_profile_helper(item, mdp, reg2, start_row + 1, done, total_rows, 1767 item_offset_fn, item_count_offset_fn); 1768 1769 // Found a null. Keep searching for a matching item, 1770 // but remember that this is an empty (unused) slot. 1771 bind(found_null); 1772 } 1773 } 1774 1775 // In the fall-through case, we found no matching item, but we 1776 // observed the item[start_row] is null. 1777 1778 // Fill in the item field and increment the count. 1779 int item_offset = in_bytes(item_offset_fn(start_row)); 1780 set_mdp_data_at(mdp, item_offset, item); 1781 int count_offset = in_bytes(item_count_offset_fn(start_row)); 1782 movl(reg2, DataLayout::counter_increment); 1783 set_mdp_data_at(mdp, count_offset, reg2); 1784 if (start_row > 0) { 1785 jmp(done); 1786 } 1787 } 1788 1789 // Example state machine code for three profile rows: 1790 // // main copy of decision tree, rooted at row[1] 1791 // if (row[0].rec == rec) { row[0].incr(); goto done; } 1792 // if (row[0].rec != nullptr) { 1793 // // inner copy of decision tree, rooted at row[1] 1794 // if (row[1].rec == rec) { row[1].incr(); goto done; } 1795 // if (row[1].rec != nullptr) { 1796 // // degenerate decision tree, rooted at row[2] 1797 // if (row[2].rec == rec) { row[2].incr(); goto done; } 1798 // if (row[2].rec != nullptr) { count.incr(); goto done; } // overflow 1799 // row[2].init(rec); goto done; 1800 // } else { 1801 // // remember row[1] is empty 1802 // if (row[2].rec == rec) { row[2].incr(); goto done; } 1803 // row[1].init(rec); goto done; 1804 // } 1805 // } else { 1806 // // remember row[0] is empty 1807 // if (row[1].rec == rec) { row[1].incr(); goto done; } 1808 // if (row[2].rec == rec) { row[2].incr(); goto done; } 1809 // row[0].init(rec); goto done; 1810 // } 1811 // done: 1812 1813 void InterpreterMacroAssembler::record_klass_in_profile(Register receiver, 1814 Register mdp, Register reg2, 1815 bool is_virtual_call) { 1816 assert(ProfileInterpreter, "must be profiling"); 1817 Label done; 1818 1819 record_klass_in_profile_helper(receiver, mdp, reg2, 0, done, is_virtual_call); 1820 1821 bind (done); 1822 } 1823 1824 void InterpreterMacroAssembler::profile_ret(Register return_bci, 1825 Register mdp) { 1826 if (ProfileInterpreter) { 1827 Label profile_continue; 1828 uint row; 1829 1830 // If no method data exists, go to profile_continue. 1831 test_method_data_pointer(mdp, profile_continue); 1832 1833 // Update the total ret count. 1834 increment_mdp_data_at(mdp, in_bytes(CounterData::count_offset())); 1835 1836 for (row = 0; row < RetData::row_limit(); row++) { 1837 Label next_test; 1838 1839 // See if return_bci is equal to bci[n]: 1840 test_mdp_data_at(mdp, 1841 in_bytes(RetData::bci_offset(row)), 1842 return_bci, noreg, 1843 next_test); 1844 1845 // return_bci is equal to bci[n]. Increment the count. 1846 increment_mdp_data_at(mdp, in_bytes(RetData::bci_count_offset(row))); 1847 1848 // The method data pointer needs to be updated to reflect the new target. 1849 update_mdp_by_offset(mdp, 1850 in_bytes(RetData::bci_displacement_offset(row))); 1851 jmp(profile_continue); 1852 bind(next_test); 1853 } 1854 1855 update_mdp_for_ret(return_bci); 1856 1857 bind(profile_continue); 1858 } 1859 } 1860 1861 1862 void InterpreterMacroAssembler::profile_null_seen(Register mdp) { 1863 if (ProfileInterpreter) { 1864 Label profile_continue; 1865 1866 // If no method data exists, go to profile_continue. 1867 test_method_data_pointer(mdp, profile_continue); 1868 1869 set_mdp_flag_at(mdp, BitData::null_seen_byte_constant()); 1870 1871 // The method data pointer needs to be updated. 1872 int mdp_delta = in_bytes(BitData::bit_data_size()); 1873 if (TypeProfileCasts) { 1874 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1875 } 1876 update_mdp_by_constant(mdp, mdp_delta); 1877 1878 bind(profile_continue); 1879 } 1880 } 1881 1882 1883 void InterpreterMacroAssembler::profile_typecheck(Register mdp, Register klass, Register reg2) { 1884 if (ProfileInterpreter) { 1885 Label profile_continue; 1886 1887 // If no method data exists, go to profile_continue. 1888 test_method_data_pointer(mdp, profile_continue); 1889 1890 // The method data pointer needs to be updated. 1891 int mdp_delta = in_bytes(BitData::bit_data_size()); 1892 if (TypeProfileCasts) { 1893 mdp_delta = in_bytes(VirtualCallData::virtual_call_data_size()); 1894 1895 // Record the object type. 1896 record_klass_in_profile(klass, mdp, reg2, false); 1897 NOT_LP64(assert(reg2 == rdi, "we know how to fix this blown reg");) 1898 NOT_LP64(restore_locals();) // Restore EDI 1899 } 1900 update_mdp_by_constant(mdp, mdp_delta); 1901 1902 bind(profile_continue); 1903 } 1904 } 1905 1906 1907 void InterpreterMacroAssembler::profile_switch_default(Register mdp) { 1908 if (ProfileInterpreter) { 1909 Label profile_continue; 1910 1911 // If no method data exists, go to profile_continue. 1912 test_method_data_pointer(mdp, profile_continue); 1913 1914 // Update the default case count 1915 increment_mdp_data_at(mdp, 1916 in_bytes(MultiBranchData::default_count_offset())); 1917 1918 // The method data pointer needs to be updated. 1919 update_mdp_by_offset(mdp, 1920 in_bytes(MultiBranchData:: 1921 default_displacement_offset())); 1922 1923 bind(profile_continue); 1924 } 1925 } 1926 1927 1928 void InterpreterMacroAssembler::profile_switch_case(Register index, 1929 Register mdp, 1930 Register reg2) { 1931 if (ProfileInterpreter) { 1932 Label profile_continue; 1933 1934 // If no method data exists, go to profile_continue. 1935 test_method_data_pointer(mdp, profile_continue); 1936 1937 // Build the base (index * per_case_size_in_bytes()) + 1938 // case_array_offset_in_bytes() 1939 movl(reg2, in_bytes(MultiBranchData::per_case_size())); 1940 imulptr(index, reg2); // XXX l ? 1941 addptr(index, in_bytes(MultiBranchData::case_array_offset())); // XXX l ? 1942 1943 // Update the case count 1944 increment_mdp_data_at(mdp, 1945 index, 1946 in_bytes(MultiBranchData::relative_count_offset())); 1947 1948 // The method data pointer needs to be updated. 1949 update_mdp_by_offset(mdp, 1950 index, 1951 in_bytes(MultiBranchData:: 1952 relative_displacement_offset())); 1953 1954 bind(profile_continue); 1955 } 1956 } 1957 1958 1959 1960 void InterpreterMacroAssembler::_interp_verify_oop(Register reg, TosState state, const char* file, int line) { 1961 if (state == atos) { 1962 MacroAssembler::_verify_oop_checked(reg, "broken oop", file, line); 1963 } 1964 } 1965 1966 void InterpreterMacroAssembler::verify_FPU(int stack_depth, TosState state) { 1967 #ifndef _LP64 1968 if ((state == ftos && UseSSE < 1) || 1969 (state == dtos && UseSSE < 2)) { 1970 MacroAssembler::verify_FPU(stack_depth); 1971 } 1972 #endif 1973 } 1974 1975 // Jump if ((*counter_addr += increment) & mask) == 0 1976 void InterpreterMacroAssembler::increment_mask_and_jump(Address counter_addr, Address mask, 1977 Register scratch, Label* where) { 1978 // This update is actually not atomic and can lose a number of updates 1979 // under heavy contention, but the alternative of using the (contended) 1980 // atomic update here penalizes profiling paths too much. 1981 movl(scratch, counter_addr); 1982 incrementl(scratch, InvocationCounter::count_increment); 1983 movl(counter_addr, scratch); 1984 andl(scratch, mask); 1985 if (where != nullptr) { 1986 jcc(Assembler::zero, *where); 1987 } 1988 } 1989 1990 void InterpreterMacroAssembler::notify_method_entry() { 1991 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to 1992 // track stack depth. If it is possible to enter interp_only_mode we add 1993 // the code to check if the event should be sent. 1994 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx); 1995 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx); 1996 if (JvmtiExport::can_post_interpreter_events()) { 1997 Label L; 1998 NOT_LP64(get_thread(rthread);) 1999 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset())); 2000 testl(rdx, rdx); 2001 jcc(Assembler::zero, L); 2002 call_VM(noreg, CAST_FROM_FN_PTR(address, 2003 InterpreterRuntime::post_method_entry)); 2004 bind(L); 2005 } 2006 2007 if (DTraceMethodProbes) { 2008 NOT_LP64(get_thread(rthread);) 2009 get_method(rarg); 2010 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_entry), 2011 rthread, rarg); 2012 } 2013 2014 // RedefineClasses() tracing support for obsolete method entry 2015 if (log_is_enabled(Trace, redefine, class, obsolete)) { 2016 NOT_LP64(get_thread(rthread);) 2017 get_method(rarg); 2018 call_VM_leaf( 2019 CAST_FROM_FN_PTR(address, SharedRuntime::rc_trace_method_entry), 2020 rthread, rarg); 2021 } 2022 } 2023 2024 2025 void InterpreterMacroAssembler::notify_method_exit( 2026 TosState state, NotifyMethodExitMode mode) { 2027 // Whenever JVMTI is interp_only_mode, method entry/exit events are sent to 2028 // track stack depth. If it is possible to enter interp_only_mode we add 2029 // the code to check if the event should be sent. 2030 Register rthread = LP64_ONLY(r15_thread) NOT_LP64(rcx); 2031 Register rarg = LP64_ONLY(c_rarg1) NOT_LP64(rbx); 2032 if (mode == NotifyJVMTI && JvmtiExport::can_post_interpreter_events()) { 2033 Label L; 2034 // Note: frame::interpreter_frame_result has a dependency on how the 2035 // method result is saved across the call to post_method_exit. If this 2036 // is changed then the interpreter_frame_result implementation will 2037 // need to be updated too. 2038 2039 // template interpreter will leave the result on the top of the stack. 2040 push(state); 2041 NOT_LP64(get_thread(rthread);) 2042 movl(rdx, Address(rthread, JavaThread::interp_only_mode_offset())); 2043 testl(rdx, rdx); 2044 jcc(Assembler::zero, L); 2045 call_VM(noreg, 2046 CAST_FROM_FN_PTR(address, InterpreterRuntime::post_method_exit)); 2047 bind(L); 2048 pop(state); 2049 } 2050 2051 if (DTraceMethodProbes) { 2052 push(state); 2053 NOT_LP64(get_thread(rthread);) 2054 get_method(rarg); 2055 call_VM_leaf(CAST_FROM_FN_PTR(address, SharedRuntime::dtrace_method_exit), 2056 rthread, rarg); 2057 pop(state); 2058 } 2059 } 2060 2061 void InterpreterMacroAssembler::load_resolved_indy_entry(Register cache, Register index) { 2062 // Get index out of bytecode pointer 2063 get_cache_index_at_bcp(index, 1, sizeof(u4)); 2064 // Get address of invokedynamic array 2065 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize)); 2066 movptr(cache, Address(cache, in_bytes(ConstantPoolCache::invokedynamic_entries_offset()))); 2067 if (is_power_of_2(sizeof(ResolvedIndyEntry))) { 2068 shll(index, log2i_exact(sizeof(ResolvedIndyEntry))); // Scale index by power of 2 2069 } else { 2070 imull(index, index, sizeof(ResolvedIndyEntry)); // Scale the index to be the entry index * sizeof(ResolvedIndyEntry) 2071 } 2072 lea(cache, Address(cache, index, Address::times_1, Array<ResolvedIndyEntry>::base_offset_in_bytes())); 2073 } 2074 2075 void InterpreterMacroAssembler::load_field_entry(Register cache, Register index, int bcp_offset) { 2076 // Get index out of bytecode pointer 2077 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize)); 2078 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2)); 2079 2080 movptr(cache, Address(cache, ConstantPoolCache::field_entries_offset())); 2081 // Take shortcut if the size is a power of 2 2082 if (is_power_of_2(sizeof(ResolvedFieldEntry))) { 2083 shll(index, log2i_exact(sizeof(ResolvedFieldEntry))); // Scale index by power of 2 2084 } else { 2085 imull(index, index, sizeof(ResolvedFieldEntry)); // Scale the index to be the entry index * sizeof(ResolvedFieldEntry) 2086 } 2087 lea(cache, Address(cache, index, Address::times_1, Array<ResolvedFieldEntry>::base_offset_in_bytes())); 2088 } 2089 2090 void InterpreterMacroAssembler::load_method_entry(Register cache, Register index, int bcp_offset) { 2091 // Get index out of bytecode pointer 2092 movptr(cache, Address(rbp, frame::interpreter_frame_cache_offset * wordSize)); 2093 get_cache_index_at_bcp(index, bcp_offset, sizeof(u2)); 2094 2095 movptr(cache, Address(cache, ConstantPoolCache::method_entries_offset())); 2096 imull(index, index, sizeof(ResolvedMethodEntry)); // Scale the index to be the entry index * sizeof(ResolvedMethodEntry) 2097 lea(cache, Address(cache, index, Address::times_1, Array<ResolvedMethodEntry>::base_offset_in_bytes())); 2098 }